Scene Intelligence for Collaborative Semantic Mapping with Mobile Robots

1. At least one non-transitory machine readable medium, including instructions for operating an autonomous mobile robot (AMR), which when executed by processing circuitry of the AMR, cause the AMR to: receive an environmental map at the AMR; cause the AMR to navigate through an environment corresponding to the environmental map; capture, at a location during navigation of the AMR through the environment, audio or video data using a sensor of the AMR; perform a classification of the audio or video data using a trained classifier; identify a coordinate of the environmental map corresponding to the location in the environment where the audio or video data was captured by the sensor during navigation of the AMR; update the environmental map to include the classification as metadata corresponding to the coordinate; communicate the updated environmental map to an edge device; and cause the AMR to access the environmental map from the edge device, the environmental map generated using federated learning, the federated learning based on data from a plurality of AMRs.

2. The at least one machine readable medium of claim 1, wherein the instructions further cause the AMR to retrieve the environmental map from the edge device prior to updating the environmental map, the environmental map including at least one classification generated by a second AMR.

3. The at least one machine readable medium of claim 1, wherein the instructions further cause the AMR to: determine, at the AMR, that an orientation or location of the AMR cannot be resolved; and in response: capture additional audio or video data; perform a second classification of the additional audio or video data using the trained classifier; determine that the second classification matches the classification; and resolve the orientation or the location of the AMR using the updated environmental map.

4. The at least one machine readable medium of claim 1, wherein the environmental map is generated using federated learning across a plurality of decentralized edge devices including the edge device.

5. The at least one machine readable medium of claim 1, wherein the environmental map includes semantic metadata generated by at least one other AMR, and wherein the instructions further cause the AMR to, prior to the use of the environmental map, determine trustworthiness of the at least one other AMR.

6. The at least one machine readable medium of claim 5, wherein to determine the trustworthiness of the at least one other AMR, the instructions further cause the AMR to (i) determine whether the at least one other AMR is authenticated or (ii) determine whether data from the at least one other AMR is accurate.

7. The at least one machine readable medium of claim 1, wherein to update the environmental map, the instructions further cause the AMR to identify a change to the coordinate based on the location.

8. The at least one machine readable medium of claim 1, wherein the environmental map includes semantic metadata generated by a static sensor within the environment.

9. The at least one machine readable medium of claim 1, wherein the sensor of the AMR includes at least one of a microphone or a camera.

10. An autonomous mobile robot (AMR) comprising: a sensor to capture, at a location during navigation of the AMR through an environment corresponding to an environmental map, audio or video data; processing circuitry; memory, including instructions, which when executed by the processing circuitry, cause the processing circuitry to: receive the environmental map at the AMR; cause the AMR to navigate through the environment; perform a classification of the audio or video data using a trained classifier; identify a coordinate of the environmental map corresponding to the location in the environment where the audio or video data was captured by the sensor during navigation of the AMR; update the environmental map to include the classification as metadata corresponding to the coordinate; communication circuitry to communicate the updated environmental map to an edge device; and cause the AMR to access the environmental map from the edge device, the environmental map generated using federated learning, the federated learning based on data from a plurality of AMRs.

11. The AMR of claim 10, wherein the communication circuitry is further to retrieve the environmental map from the edge device prior to updating the environmental map, the environmental map including at least one classification generated by a second AMR.

12. The AMR of claim 11, wherein the second AMR was made by a different manufacturer than the AMR.

13. The AMR of claim 10, wherein the instructions further cause the processing circuitry to: determine, at the AMR, that an orientation or location of the AMR cannot be resolved; and in response: capture additional audio or video data; perform a second classification of the additional audio or video data using the trained classifier; determine that the second classification matches the classification; and resolve the orientation or the location of the AMR using the updated environmental map.

14. The AMR of claim 10, wherein the environmental map is generated using federated learning across a plurality of decentralized edge devices including the edge device.

15. The AMR of claim 10, wherein the environmental map includes semantic metadata generated by at least one other AMR, and wherein the instructions further cause the AMR to, prior to the use of the environmental map, determine trustworthiness of the at least one other AMR.

16. The AMR of claim 15, wherein to determine the trustworthiness of the at least one other AMR, the instructions further cause the processing circuitry to (i) determine whether the at least one other AMR is authenticated or (ii) determine whether data from the at least one other AMR is accurate.

17. The AMR of claim 11, wherein to update the environmental map, the instructions further cause the processing circuitry to identify a change to the coordinate based on the location.

18. The AMR of claim 10, wherein the environmental map includes semantic metadata generated by a static sensor within the environment.

19. The AMR of claim 10, wherein the sensor includes at least one of a microphone or a camera.

20. An apparatus comprising: means for receiving an environmental map at an autonomous mobile robot (AMR); means for causing the AMR to navigate through an environment corresponding to the environmental map; means for capturing, at a location during navigation of the AMR through the environment, audio or video data; means for classifying the audio or video data into a classification; means for identifying a coordinate of the environmental map corresponding to the location in the environment where the audio or video data was captured during navigation of the AMR; means for updating the environmental map to include the classification as metadata corresponding to the coordinate; means for uploading the updated environmental map to an edge device; and cause the AMR to access the environmental map from the edge device, the environmental map generated using federated learning, the federated learning based on data from a plurality of AMRs.

21. The apparatus of claim 20, further comprising means for retrieving the environmental map from the edge device prior to updating the environmental map, the environmental map including at least one class generated by a second apparatus.

22. The apparatus of claim 21, wherein the apparatus is an autonomous mobile robot (AMR) and wherein the second apparatus is an AMR that was made by a different manufacturer than the apparatus.

23. The apparatus of claim 20, wherein the environmental map includes semantic metadata generated by a static sensor within the environment.

24. The apparatus of claim 20, wherein the environmental map is generated using federated learning across a plurality of decentralized edge devices including the edge device.